MobilityGen: Diffusion AI simulates human travel patterns over weeks

A team led by Ye Hong from ETH Zurich has developed MobilityGen, a diffusion-based deep generative framework that models human mobility behavior with unprecedented fidelity. Building on the activity-based view of daily travel, the system generates multi-attribute sequences of activities, locations, times, and travel modes over periods from days to weeks, at large spatial scales. By linking behavioral attributes with environmental context, MobilityGen reproduces key empirical patterns — including scaling laws for location visits, time allocation across activities, and the coupled evolution of mode choice and destination selection. It captures spatio-temporal variability and produces diverse, plausible mobility patterns consistent with the built environment.

Beyond standard validation, MobilityGen opens doors to analyses that were previously difficult with earlier models. It can measure how access to urban space varies across different travel modes (e.g., car vs. transit vs. walking) and how co-presence dynamics shape social exposure and segregation. The model is evaluated against real-world data and demonstrates strong performance in replicating aggregate mobility statistics. This work, published on arXiv (2510.06473), sits at the intersection of physics, AI, and social networks, offering a data-driven basis for transport planning, sustainable urban design, and public health interventions.